Development of Performance Assessment MethodologiesDevelopment of
Performance Assessment Methodologies
Development of Performance Assessment Methodologies
André Rübel Dirk-Alexander Becker Eckard Fein Alice Ionescu Thomas
Lauke Jörg Mönig Ulrich Noseck Anke Schneider Sabine Spießl Jens
Wolf
October 2010
Remark:
This report was prepared under contract no. 02 E 10276 with the
Federal Ministry of Economics and Technology (BMWi).
The work was conducted by the Gesellschaft für Anlagen- und
Reaktorsicherheit (GRS) mbH.
The authors are responsible for the content of this report.
GRS - 259 ISBN 978-3-939355-34-2
I
Abstract
The project PAMINA (Performance Assessment methodologies in
application to guide
the development of the safety case) was conducted from 2007 to 2009
on the Europe-
an level to improve and harmonise integrated performance assessment
methodologies
and tools for various disposal concepts of long-lived radioactive
waste and spent nu-
clear fuel in different deep geological environments. The following
report presents the
contributions of the GRS Braunschweig to the PAMINA project
comprising the following
topics:
− Overview of methodologies, tools and experiences:
A comprehensive review is presented from the point of view of the
implementer to
assess the state of the art of the methodologies and approaches
needed for the
safety assessment of geological repositories for the German
national programme,
and to distil the lessons learned from the rich experience
accumulated in their de-
velopment and application.
− Treatment of uncertainty in integrated performance
assessment:
A protocol is presented that helps to determine adequate
probability density func-
tion for parameter values to deal with parameter uncertainties in
probabilistic safety
analyses.
Different sensitivity analysis methods have been tested for a
performance
assessment model for a high-level waste repository in rock salt.
These methods
include variance based FAST and EFAST methods.
− Use of safety indicators and performance indicators:
Different safety and performance indicators have been tested for
high level waste
repositories in rock salt and clay formations. For the repository
in clay sensitivity
analysis methods have been tested to gain deeper insight into the
performance of
subsystems.
− Relevance of sophisticated approaches in practical cases:
The performance assessment approach to three selected near-field
processes in a
repository in salt have been tested for their suitability. The
three processes are
convergence of salt, brine intrusion into a backfilled drift and
convective driven
transport. Additionally, the relevance of the complexity of
modelling for the far field
of a repository in salt has been assessed.
II
III
2.1 Current status of the German national context regarding
repository
projects
......................................................................................................
5
2.2.1 Background/Introduction
............................................................................
7
2.2.3 Treatment in the safety
case....................................................................
10
2.2.3.1 Methodology
............................................................................................
11
2.2.3.5 On-going work and future evolution
......................................................... 16
2.2.4 Lessons learnt
.........................................................................................
17
2.3.1 Background/Introduction
..........................................................................
18
2.3.3 Treatment in the safety
case....................................................................
21
2.3.3.1 Methodology
............................................................................................
21
2.3.3.5 On-going work and future evolution
......................................................... 29
2.3.4 Lessons learnt
.........................................................................................
29
2.4.1 Background
.............................................................................................
30
2.4.3.1 Safety concept
.........................................................................................
32
2.4.4.1 Normal evolution scenario
.......................................................................
34
2.4.4.2 Altered evolution scenarios
......................................................................
38
2.4.5 Lessons learnt and outlook
......................................................................
39
2.5 Modelling strategy
...................................................................................
40
2.5.2 Regulatory requirements and provisions
.................................................. 40
2.5.3 Key terms and concepts
..........................................................................
41
2.5.4 Treatment in the Safety Case
..................................................................
41
2.5.4.1 Methodology
............................................................................................
41
2.5.4.4 On-going work and future evolution
......................................................... 50
2.5.5 Lessons learned
......................................................................................
50
2.6 Sensitivity analysis
..................................................................................
51
2.6.3 Treatment in the safety
case....................................................................
53
2.6.3.1 Methodology
............................................................................................
53
2.6.3.5 On-going work and future evolution
......................................................... 63
2.6.4 Lessons learnt
.........................................................................................
63
2.7.2 Regulatory requirements and provisions
.................................................. 65
2.7.3 Key terms and concepts
..........................................................................
65
2.7.4 Treatment in the Safety Case
..................................................................
66
2.7.4.1 Methodology
............................................................................................
66
2.7.4.5 On-going work and future evolution
......................................................... 71
2.7.5 Lessons learnt
.........................................................................................
71
2.8 Human intrusion
......................................................................................
72
2.8.2 Key terms and concepts
..........................................................................
72
2.8.3 Treatment in the Safety Case
..................................................................
73
2.8.3.1 Methodology
............................................................................................
73
2.9 Criteria for input and data selection
......................................................... 79
2.9.1 Background
.............................................................................................
80
2.9.4 Treatment in the safety
case....................................................................
82
2.9.4.1 Methodology
............................................................................................
82
2.9.4.4 On-going work and future evolution
....................................................... 100
2.9.5 Lessons learnt
.......................................................................................
100
in natural claystone
................................................................................
101
3.1 Protocol for assessing parameter uncertainty
........................................ 109
3.1.1 General procedure and practical considerations
.................................... 110
3.1.2 Selection and assessment of a knowledge
base.................................... 111
3.1.2.1 Assessment of the quality level of information
....................................... 111
3.1.2.2 Amalgamation of different sources into one data set
............................. 113
VI
3.1.3 Assessment of parameter uncertainty
.................................................... 116
3.1.3.1 Consistency
...........................................................................................
116
3.1.3.3 Determination of a PDF from given data
................................................ 118
3.1.4 Algorithmic description of PDF generation
............................................. 121
3.2 Application of different sensitivity analysis methods to a
PA
model for a repository in rock salt
..........................................................
124
3.2.1 Modification of the EMOS statistic module for the use of
FAST
methods
................................................................................................
126
3.2.3.2 Time-dependent analysis
.......................................................................
139
4 Safety indicators and performance indicators
.................................. 151
4.1 Repository in salt
...................................................................................
151
4.1.1 Reference concept and scenario information
......................................... 152
4.1.2 Safety indicators
....................................................................................
157
4.1.2.2 Radiotoxicity concentration in biosphere water
...................................... 161
4.1.2.3 Power density in groundwater
................................................................
163
4.1.2.4 Radiotoxicity flux to/from the geosphere
................................................ 169
4.1.2.5 Normalised safety indicators
..................................................................
171
4.1.2.6 Robustness of safety indicators in case of radionuclide
release ............ 173
4.1.3 Indicators based on risk
.........................................................................
176
4.1.3.1 Reference values for indicators based on risk
....................................... 180
4.1.3.2 Calculation of risk
..................................................................................
181
4.1.4 Performance indicators
..........................................................................
183
4.1.4.1 Compartment structure
..........................................................................
184
VII
4.1.4.4 Integrated radiotoxicity fluxes from compartments
................................. 188
4.1.5 Summary
...............................................................................................
189
4.2.1 Test case
...............................................................................................
194
4.2.2.1 Enhancements of the CLAYPOS
module............................................... 197
4.2.2.2 Radiotoxicity inventories/fluxes in/from different
compartments ............. 204
4.2.3 Results from the probabilistic simulations for the dose as
safety
indicator
.................................................................................................
212
4.2.3.2 Analysis of the maximum dose rate
....................................................... 214
4.2.3.3 Time-dependent analysis
.......................................................................
219
4.2.4 Results from the probabilistic simulations for radiotoxicity
fluxes and
inventories as performance indicators
................................................... 229
4.2.4.1 EFAST analysis
.....................................................................................
229
4.3.1 Safety Indicators
....................................................................................
246
5 Relevance of sophisticated approaches in practical cases
............. 255
5.1 Testing of the PA approaches for selected near-field processes
in a
repository in salt
....................................................................................
255
5.1.1.1 Definition of the test cases
.....................................................................
257
5.1.1.2 Results
..................................................................................................
260
5.1.1.3 Conclusions
...........................................................................................
277
5.1.2 Benchmark on brine intrusion into a backfilled drift
................................ 278
5.1.2.1 Test case
...............................................................................................
279
5.1.3.1 Test case
...............................................................................................
298
5.1.3.2 Results
..................................................................................................
300
5.1.3.3 Conclusions
...........................................................................................
305
5.2 Relevance of the complexity of modelling for the far field of
a
repository in salt
....................................................................................
306
5.2.1 Test case
...............................................................................................
306
5.2.2.2 2D transport model
................................................................................
310
5.2.2.3 Abstraction to 1D-model
........................................................................
317
5.2.3 Conclusions
...........................................................................................
325
5.3 Coupling of the transport code r3t with the geochemical
code
Phreeqc
.................................................................................................
327
5.3.2 Salt water intrusion into a Ca-HCO3-water column
................................ 331
5.3.3 Conclusion
.............................................................................................
332
6 References
...........................................................................................
333
A Annex: Example files for the use of the FAST method in EMOS
...... 343
List of figures
.......................................................................................
355
List of tables
........................................................................................
365
1 Introduction
A comprehensive set of arguments and analyses – represented in a
safety case – is
needed to justify that geological disposal of long-lived
radioactive waste and spent nu-
clear fuel is safe. One pillar of the safety case is the integrated
performance assess-
ment of the repository by numerical methods. This Performance
Assessment requires a
powerful and qualified instrument. The approach used for the
performance assessment
must meet national regulations on the one hand and should be
internationally accepted
on the other hand. Further it must be continuously adapted to the
state of the art of sci-
ence and technology. Computer codes used for the assessments must
be tested and
verified and be designed for the prerequisites of a real waste
repository system.
On the European level the project PAMINA (Performance Assessment
methodologies
in application to guide the development of the safety case) was
conducted from 2007 to
2009 to improve and harmonise integrated performance assessment
methodologies
and tools for various disposal concepts of long-lived radioactive
waste and spent nu-
clear fuel in different deep geological environments. PAMINA aimed
at providing a
sound methodological and scientific basis for demonstrating the
safety of deep geolog-
ical disposal of such wastes, that will be of value to all national
radioactive waste man-
agement programmes, regardless of waste type, repository design,
and stage, that has
been reached in PA and safety case development.
The following report presents the different contributions of the
GRS Braunschweig to
the different tasks of the PAMINA project. On a national level,
this work was co-funded
by the Federal Ministry of Economics and Technology (BMWi). Some of
the work de-
scribed in the following is presented in a similar way in public
PAMINA reports given at
the according sections.
The following chapter gives for selected topics an overview of PA
methodologies, tools
and experiences for the German national programme from the point of
view of the im-
plementer. The other chapters (3 to 5) present methodological
advancements achieved
in different areas, which are
− the treatment and management of uncertainty during PA and safety
case develop-
ment (chapter 3),
− the use of safety and performance indicators for repositories in
salt and clay (chap-
ter 4) and finally,
2
− the improvement of methods and tools regarding process
understanding and con-
ceptualization and the determination of needs for implementing more
sophisticated
modelling approaches in PA (chapter 5).
3
2 Overview of methodologies, tools and experiences
During the last decades a very large body of experience regarding
safety assessment
of geological repositories of radioactive waste has been generated,
both in European
countries and outside of Europe. This experience provides a firm
basis for future steps
in national development programmes. In parallel with development
activities, a growing
number of formal evaluation processes, including regulatory
processes, have been and
are being carried out, generating important guidance for future
work. A significant part
of these efforts has been realised under the auspices, and in the
framework, of the
programmes of international organisations. A comprehensive review
was performed in
the project PAMINA with the objective to assess the state of the
art of the methodolo-
gies and approaches needed for the safety assessment of geological
repositories, and
to distil the lessons learned from the rich experience accumulated
in their development
and application. The following issues have been addressed:
1. Safety indicators and performance/function indicators
2. Safety functions
5. Assessment strategy - Safety Approach
6. Evolution of the repository system
7. Modelling strategy
8. Sensitivity analysis
11. Criteria for input and data selection
The following nine sections of this chapter give the review work of
the GRS Braun-
schweig addressing the German radioactive waste disposal programme
from the view-
point of the implementer. The overall results of the PAMINA review
are presented in
the RTDC1 deliverables /PAM 06, PAM 09, PAM 10/. All the topics
share a common
structure including the following areas:
4
− Background/Introduction
− Methodology
− Lessons learnt
Not all areas are addressed in each of the topic. The statements
towards the regulatory
requirements are similar for all the topics and are summarised in
the following. Addi-
tional comments in each of the topics chapters towards the
regulatory requirements are
only given if deviant or supplementary comments have been given to
that specific topic.
The German Atomic Energy Act merely requires the safe disposal of
radioactive waste.
There is an old German guideline (“safety criteria for the final
disposal of radioactive
wastes in a mine”), originating from 1983, which is formally still
valid /BMI 83/. Con-
cerning long-term safety, it simply requires that “even after
decommissioning radionu-
clides that could reach the biosphere in consequence of
non-excludable transport pro-
cesses from a sealed repository must not lead to individual doses
exceeding the value
given in the Radiation Protection Ordinance”. This value is 0.3
mSv/yr and is valid for
all nuclear facilities. A supplementary regulation from 1988
defines the time frame for
which the individual dose rate should be evaluated as 10 000 years.
The consideration
of other safety indicators is not required, nor are probabilistic
criteria defined. There is,
however, a consensus in Germany that the mentioned guideline is
outdated and should
be revised soon. A first draft for a new version, proposed by GRS,
is currently under in-
tense discussion. It requires the consideration of six indicators
with fixed reference val-
ues as well as a probabilistic analysis. This paper is, however, a
controversial matter
and will be essentially changed before being accepted by the
authorities. Therefore, it
5
is not presented here. Nevertheless, it can be said that the future
guideline is very likely
to contain the following regulations:
− the calculated individual effective dose rate must not exceed the
reference value of
0.1 mSv/yr,
− several additional safety indicators have to be calculated,
− the time frame for which safety has to be proven is 1 million
years or more.
Besides the review performed within the PAMINA project the GRS also
followed the
German network for research on the actinide migration in natural
claystone from the
long-term safety assessment point of view. This view is given in
the final section 2.10 of
this chapter.
2.1 Current status of the German national context regarding
repository
projects
According to the Atomic Energy Act /ATG 85/ the German Federal
Government has to
ensure the safe disposal of radioactive waste by providing
repositories. The legal basis
for licensing of federal installations for the safekeeping and
final disposal of radioactive
waste is the "Plan Approval Procedure" required by the Atomic
Energy Act. Radioactive
waste disposal policy in Germany is based on the decision that all
types of radioactive
waste are to be disposed of in deep geological formations. The
currently valid safety
criteria for the final disposal of radioactive waste in a mine
dates from 1983 /BMI 83/.
Since then, regulatory expectations have advanced, now reflecting
the international
standards set out by ICRP /ICRP 98/, NEA /NEA 04/ and IAEA /IAEA
06/. On this ac-
count, GRS proposed “Safety requirements for the disposal of high
active wastes in
deep geological formations” /BAL 07/ on behalf of BMU (Federal
Ministry for the Envi-
ronment, Nature Conservation and Nuclear Safety), which is expected
to serve as a
sound basis for a new regulation. The BMU is presently elaborating
the final version of
the Safety Requirements. A draft version of the Safety Requirements
was presented in
November 2008 to the public /BMU 08/.
Prior to 1980 the former iron ore mine Konrad was selected as a
site for disposal of
short-lived and long-lived radioactive waste with negligible heat
generation and the salt
dome at Gorleben as a site for the disposal of all types of
radioactive waste. In the for-
6
mer German Democratic Republic short-lived low- and
intermediate-level radioactive
waste was disposed of in the Morsleben repository, a former rock
salt and potash mine.
The Konrad repository had been licensed in May 2002. All suits that
were filed against
it were rejected by the competent court in 2006. Complaints against
the courts decision
were definitely rejected by the Federal Administrative Court in
April 2007. Following
necessary planning adjustments the former iron ore mine will be
converted into a re-
pository for all kinds of radioactive waste with negligible heat
generation by the end of
2013.
The disused salt and potash mine Morsleben (ERAM), located in the
Federal State
of Saxony-Anhalt, has been in operation since 1971 as a repository
for short-lived low-
and intermediate-level radioactive waste. Disposal was terminated
in 1998. A waste
volume of about 37 000 m3 has been disposed of with a total
activity of approx.
4.5·1014 Bq. Since 1990, the Morsleben facility has the status of a
federal repository.
The license for operating the repository originates from the former
German Democratic
Republic and do not include the license for the closure of the
repository. Therefore, ac-
cording to the German Atomic Energy Act /ATG 85/ a license
application for the closure
of the repository is being prepared by BfS (Federal Office for
Radiation Protection).
The Gorleben salt dome in the north-east of Lower Saxony has been
investigated for
its suitability to host a repository for all types of solid and
solidified radioactive waste for
several decades. However, after the licensing of the Konrad
repository the focus is
mainly on heat generating radioactive waste originating from
reprocessing and spent
fuel elements. The exploration of the Gorleben salt dome was
interrupted on 1st Octo-
ber 2000 according to a moratorium of up to 10 years.
The German radioactive waste management and disposal concept as
well as the site
selection process are still under discussion. In terms of the site
selection process, a re-
spective concept from BMU was suggested in 2006. This concept
includes the exami-
nation whether site alternatives exist in addition to Gorleben,
which let expect or pos-
sess a higher level of safety /GAB 08/.
The current R&D concept focuses on all types of host rocks,
prioritised in the following
order: rock salt, argillaceous rock, crystalline rock. Concerning
rock salt, which has
been the preferred option in Germany for several decades, the
technical and engineer-
ing know how as well as the scientific expertise are considered
well advanced and are
7
now available for the conceptual design of a high level waste
repository. During the last
10 to 15 years suitable analytical tools have been continuously
developed according to
the world wide advancing state-of-the-art. They are ready to be
tested and applied at
appropriate and concrete cases. For repositories in argillaceous
and crystalline rock
R&D work focussing mainly on mechanical and hydraulic
properties of the engineered
and the geological barriers has been performed during the last
decade. System models
for an integrated safety assessment are available for both
formations.
2.2 Safety indicators and performance/function indicators
2.2.1 Background/Introduction
Although, of course, measures for quantifying the results of
performance assessment
calculations, mainly dose and risk, were always in use, it is a
relatively new concept to
improve the understanding of the system and to support the safety
case by using com-
plementary indicators. Such indicators are calculable quantities
resulting from a PA
calculation. While safety indicators aim at providing a
quantitative criterion for the over-
all safety of a repository system, other indicators are calculated
and presented to show
the functioning of the system or specific components. They are
sometimes called ‘per-
formance indicators’ or ‘function indicators’, but they differ,
with respect to goals and in-
tention, from what SKB calls ‘safety function indicators’. It is
therefore suggested, in or-
der to avoid confusion, to use the term ‘function indicator’ only
in the latter sense as a
short form. In this paper, the term ‘performance indicator’ is
used.
In former German safety assessment studies, the only safety
indicator used was the
individual ingestion dose per year, compared to a regulatory limit.
The SPIN project
/BEC 03/ was initiated by a new way of thinking, based on the
awareness that the ro-
bustness of the safety case could be improved by using more than
one safety indicator,
as well as performance indicators. Several safety and performance
indicators were
tested in SPIN, using four national granite studies as
examples.
In 2004, a detailed performance assessment for the Morsleben LAW
repository
(ERAM), which is installed in a former salt mine, was performed.
The safety indicators
defined in SPIN, as well as some performance indicators, were
successfully applied to
support the safety statement. It has become clear in this exercise
that a rock salt re-
pository requires performance indicators that differ from those
used for granite, while
8
safety indicators, though possibly depending on local reference
values, are independ-
ent of the site and formation type.
The concepts and understanding of safety and performance indicators
have further
evolved since the end of SPIN. Presently, a new study for a HLW/SF
repository in salt,
called ISIBEL, is being made. Several safety and performance
indicators were or will
be calculated and compared with one another. This is done in
parallel to PAMINA and
the new concepts and ideas developed.
2.2.2 Key terms and concepts
In the following, the concept of safety and performance indicators
as it is understood by
GRS (Braunschweig) is described. Since the subject is under intense
discussion in
Germany at present, the following should neither be seen as ‘the
German standpoint’,
nor should it be regarded as final.
Safety indicators
Repositories for radioactive waste must be proven to be safe in the
long-term. But what
does that mean? A very general definition of repository safety can
be given in the fol-
lowing way:
A repository is safe if it does not significantly change or disturb
the natural evolution of
the environment outside a narrowly limited area of influence.
Safety, in this sense, cannot be reduced to one single aspect like
human health, but
comprises a nearly unlimited variety of protection goals like water
quality, air quality,
protection of species, etc. There can, of course, be overlap
between such protection
goals, or one goal can completely include another one, but the
often-heard statement
that protection of man comprises all other protection goals cannot
be proven.
A numerical calculation of the dissemination of radionuclides from
a repository yields,
in general, radionuclide fluxes. These results are per se not
suitable for assessing the
long-term safety of the repository, as they give no information
about whether or not the
repository can be considered ‘safe’ as defined above. It is
necessary to convert the re-
sults into some safety-related measure, or safety measure.
‘Safety-related’ means that
the safety measure should quantify a specific aspect of repository
safety.
9
The word ‘significantly’ in the definition above does allow a
certain influence of the re-
pository on the environment if it is very small and negligible in
comparison with natural
influences. If safety with respect to some specific aspect is to be
assessed using a
safety measure, it is necessary to quantify a reference value as
the limit of acceptability
with respect to the safety aspect under consideration. Reference
values should be
proven to maintain the protection goal.
It is possible that different safety aspects (or protection goals)
can be quantified with
the same safety measure, using different reference values.
Therefore, only the combi-
nation of a safety measure and a suitable, safety related reference
value, both related
to the same protection goal, is appropriate to give an indication
of safety of the reposi-
tory and is called a safety indicator. A safety indicator should
always take account of
the effects of all radionuclides in the repository.
There are two kinds of safety indicators. Those of the first kind
are calculated for spe-
cific scenarios and the results can be compared in order to assess
the consequences
of different scenarios. Safety indicators of the second kind,
however, are summed up
over all relevant scenarios, each weighted by its probability. Such
indicators are prefer-
ably calculated in terms of risk. They can be compared with risks
from daily life or from
natural sources like earthquakes, meteorite impacts, etc. The main
problem with risk
indicators is that scenario probabilities can, in most cases, only
be roughly estimated.
For performing a safety assessment it is always necessary to use at
least one safety
indicator. The technique mostly applied in the past is to calculate
the time-evolution of
the annual ingestion dose to an individual or a group and to
compare it with a regulato-
ry limit. The protection goal underlying this safety indicator is
human health and the
reference value was, though fixed by a regulatory rule, originally
derived from the de-
mand to be negligible compared to the natural background. In
Germany, a value of
0.3 mSv/yr has been used so far. This safety indicator is widely
used and refers to a ra-
ther universal protection goal, but it depends on more or less
uncertain assumptions
about the geosphere and biosphere. Moreover, it could suggest
covering all aspects of
safety, while actually it does not. Therefore, it is regarded
increasingly necessary to
consider additional safety indicators.
Performance indicators
Safety indicators are a good means to assess the overall safety of
a repository system,
but they do not yield detailed information about the functioning of
the system. Such in-
formation, however, can be very helpful or even necessary in the
process of concept
development. It can be gained by using performance
indicators.
A performance indicator is a calculable measure for the performance
of parts of the
system. These parts, which are called compartments, can be things
like single barriers,
groups of barriers, emplacement fields, the complete near field, or
even the total sys-
tem. Compartments can include others. The compartment structure to
be used for a
specific repository system should be a sensibly simplified image of
the real system
structure and depends on the type of the repository.
Performance indicators should illustrate how the repository works.
Radionuclide fluxes
between or concentrations in the compartments, e. g., show how and
where the radio-
nuclides are retained during the transport through the system. The
time-evolution of a
performance indicator should be calculated and compared for
different locations, but a
comparison with an absolute value is normally not necessary.
Whereas a safety indicator always requires considering of all
relevant radionuclides in
order to derive a safety statement, a performance indicator can be
calculated for a sin-
gle radionuclide, a group of radionuclides or the total
radionuclide spectrum, depending
on what is to be demonstrated. In this way it is possible, e. g.,
to compare the system
performance for sorbing and non-sorbing species, or for the uranium
and the thorium
chain.
2.2.3 Treatment in the safety case
This section describes which indicators have been used by GRS in
the past, and why.
It is pointed out how the indicators have been calculated and
interpreted and which ref-
erence values were used.
2.2.3.1 Methodology
Safety indicators
According to the regulations mentioned above, in all German studies
made before
2000, only the individual effective dose rate was calculated and
compared with the limit
of 0.3 mSv/yr, normally for different concepts or different
scenarios. Additional numeri-
cal investigations were, in some cases, performed in order to
explain the results, but
not to derive independent safety statements. In contrast to what
the valid guidelines re-
quire, however, the calculations were always executed over a model
time of at least 1
million years.
The SPIN project (2000 – 2002) has triggered a new view of the
problem. The three
safety indicators identified in SPIN to be useful have been applied
in two recent studies
for real sites:
− ERAM: The long-term safety assessment study for the LAW
repository in rock salt
near Morsleben,
− Asse: The long-term safety assessment study for the experimental
LAW/MAW re-
pository Asse in rock salt near Wolfenbüttel.
Moreover, five of the six indicators defined in the GRS proposal
for a new guideline
have recently been calculated in the ISIBEL study which considers a
generic HAW re-
pository in rock salt. This, however, is a running project, and the
indicators themselves
are still under discussion at GRS. Therefore, the results and
findings of this exercise
are not presented here.
In the following, the application of the SPIN safety indicators in
the ERAM study is ex-
plained more detailed.
The primary safety indicator evaluated in the study is, according
to the regulations
mentioned above, still the effective dose rate to an adult human
individual, in combina-
tion with the regulatory reference value of 0.3 mSv/yr. It has been
calculated as a func-
tion of time over 1 million years, using standardised biosphere
dose conversion factors.
12
These dose conversion factors have been defined by GSF considering
a number of
typical exposure paths, which comprise:
− ingestion of drinking water,
− inhalation of contaminated particles,
− exposure by external radiation.
Since these paths refer to the present human population, the dose
conversions factors
are increasingly uncertain for longer time frames.
There was no freedom about the reference value, but since it is
about 10 % of the natu-
ral radiation exposure, the repository is considered to be safe if
the additional radiation
exposure originating from it remains below this limit. For the ERAM
reference scenario,
the maximum dose rate is more than three orders of magnitude below
the reference
value.
Two more safety indicators have been considered. The radiotoxicity
concentration in
the aquifer has been calculated using the ingestion dose
coefficients by ICRP. This
measure is more robust than the dose rate because it is independent
of the biosphere,
though it is still based on the radiosensitivity of present-day
humans. There is no “offi-
cial” reference value for this measure, but it is rather easy to
determine one. Waters
that have been drunk by humans for hundreds of years without
causing harm can be
considered radiologically safe. There are a lot of data about
concentrations of radionu-
clides in German drinking waters, and a typical radiotoxicity
concentration of
7.7 µSv/m³ could be derived. With this reference value the
radiotoxicity concentration
becomes a proper safety indicator. It has been found that for the
ERAM reference sce-
nario the maximum radioxicity concentration in the aquifer is a
little more than three or-
ders of magnitude below this reference value.
The third safety indicator considered is based on the radiotoxicity
flux from the reposi-
tory. This is an even more robust measure than the aquifer
concentration because it is
independent of the geosphere, which could be influenced by ice ages
etc. The problem
13
with this measure is to find a clearly safety-related reference
value. Two different pos-
sibilities were discussed. One is the natural radiotoxicity flux in
a river near the reposi-
tory, which is likely to finally collect all radionuclides released
from there. The other
possibility is the natural flux of raditoxicity in the groundwater
near the repository. It was
found that the second value was about three orders of magnitude
lower than the first
one. This is an example for the argument that one single safety
measure can yield dif-
ferent and independent safety indicators if compared with different
reference values. If
the first value is used, the safety statement will be, “there is no
significant influence on
the river”, which could be relevant for the river fauna and is
clearly a safety aspect. If,
however, the groundwater flux is used as reference value, the
safety statement will be,
“there is no significant influence on the groundwater”, which is a
different and probably
more relevant safety aspect. By this reason, and because the value
is lower, it was de-
cided only to consider the natural radiotoxicity flux in
groundwater as reference value,
though it was harder to determine and is considered less robust. It
was found to be
0.2 Sv/yr. For the ERAM reference scenario the maximum
radiotoxicity flux from the
repository is a bit more than three orders of magnitude below this
reference value.
Performance indicators
In order to investigate the functioning of the repository system in
detail, several perfor-
mance indicators have been calculated for the ERAM reference
scenario. The com-
partment structure used for this purpose is based on the model
structure which is a
strong simplification of the real mine structure. There are three
sealed emplacement
areas, two non-sealed emplacement areas and a number of voids that
have not been
used for emplacement purposes and are called ‘residual mine’.
Depending on the spe-
cific requirements of the investigations, the performance
indicators have been calculat-
ed for slightly different compartment structures, sometimes merging
the non-sealed
emplacement fields together with the residual mine, sometimes not.
It has become
clear that, unlike a granite repository as considered in SPIN, a
rock salt repository, es-
pecially if erected in an abandoned production mine, does not allow
a unique and hier-
archical compartment structure.
In order to show the dissemination of radionuclides within the
mine, the concentration
of radiotoxicity in the different compartments has been calculated
as a function of time.
To distinguish between the influences of the different emplacement
fields, three differ-
ent investigations were performed, one considering the total
inventory, one considering
only the inventory of the sealed emplacement areas, and one
considering only the in-
14
ventory of the non-sealed emplacement areas. It could be showed
that the sealed em-
placement areas, though the seals are assumed to lose their
effectiveness after about
20 000 years, still contain 90 % of that part of their inventory
that has not decayed after
1 million years. Even the non-sealed emplacement areas hold the
main part of their in-
ventory for about 100 000 years.
As an additional performance indicator the integrated radiotoxicity
flux from the com-
partments was calculated as a function of time, each normalised to
the initial inventory
of the appropriate compartment. As already detected in SPIN, this
is a very illustrative
indicator since the time curves reach asymptotic values and the
comparison of these
shows how much of the inventory is finally retained in each
compartment. The results
show that a part of less than 0.1 % of the inventory of the sealed
emplacement areas
leaves these and even from the worst of the non-sealed emplacement
areas only 10 %
of its inventory can escape. A part of 10-5 of the total inventory
leaves the repository
system and reaches the biosphere.
2.2.3.2 Related topics
The issue of safety and performance indicators is related to a
number of other topics:
− assessment strategy,
− safety approach,
− safety functions,
− biosphere,
− sensitivity analysis.
2.2.3.3 Databases and tools
Reference data are of high importance for safety indicators and
should be taken from
environmental measurements, biological investigations, etc. Some of
the available data
15
needed for determination of reference values are rather incomplete
and uncertain. This
problem might make it hard to apply or even test some promising
indicators.
The tools needed for calculating safety and performance indicators
are the same that
are being used for conventional performance assessment
calculations, with a few slight
modifications or add-ons.
2.2.3.4 Application and experience
In the ERAM study three safety indicators were applied as described
in section 4.1.
The time-curves are similar in shape because they have been derived
from the same
calculations, but nevertheless yield independent safety statements
since the reference
values have been determined completely independently and with
totally different as-
sumptions. It is interesting to see that even so all three safety
indicators yield nearly
exactly the same gap of about three orders of magnitude between the
maximum and
the reference value. This is clearly a coincidence but it shows a
certain robustness of
the safety assessment. For the ERAM reference case, the results are
shown in figure
2.1 in units relative to the respective reference value. In this
representation the three
curves are very close to each other.
Time [years]
R el
at iv
e U
ni ts
10-6
10-5
10-4
10-3
10-2
10-1
100
Radiotoxicity concentration in groundwater / 7.7 µSv/m³
Radiotoxicity flow in groundwater / 0.2 Sv/a Individual dose rate /
0.3 mSv/a
Fig. 2.1 Three safety indicators, calculated for the ERAM reference
case
16
A very illustrative performance indicator is the time-integrated
radiotoxicity flow from
different compartments of the repository, related to the initial
inventory of the compart-
ment. The curves finally reach stationary values which show how
much of the initial in-
ventory leaves the compartment. For the ERAM case, this indicator
has been calculat-
ed for five compartments, three of them being separated emplacement
areas plus the
complete mine and the total system including the geosphere. The
results are shown in
figure 2.2. It can bee seen that even the worst (and non-sealed)
emplacement area,
which is not designed to retain anything at all, nevertheless
retains nearly 90 % of its
inventory and the total system releases only about ten millionths
of the initial radiotoxi-
city.
10-6
10-5
10-4
10-3
10-2
10-1
100
WSF (emplcement area) ZT (emplacement area) NF (emplacement area)
Mine Total system
Fig. 2.2 Time-integrated radiotoxicity fluxes from different
compartments of the
ERAM repository (reference case; each curve is related to the
initial in-
ventory of the respective compartment)
2.2.3.5 On-going work and future evolution
Currently, different safety and performance indicators are being
calculated within the
new ISIBEL study for a HLW/SF repository in rock salt. Within
PAMINA a wider variety
of indicators including those of the risk type is tested. It is
also planned to perform
17
probabilistic analyses in order to identify the specific
sensitivities of different safety and
performance indicators.
2.2.4 Lessons learnt
The application of different safety indicators does only make sense
if they aim at differ-
ent safety aspects and provide different and independent safety
statements. A safety
statement depends not only on the safety measure but also on the
reference value. For
a safety indicator to be robust it is necessary that neither the
safety measure nor the
reference value depend on uncertain data or assumptions. Therefore,
the radiotoxicity
flux from the repository can only be considered robust and adequate
for long time-
frames if combined with a robust and safety-related reference
value, which is not easy
to find. Establishing of reference values is a very important and
sometimes difficult
task. A good reference value should be provably safe and valid for
a long or at least
well-known time frame. Reference values can be global or
site-specific. A safety indica-
tor can never be better than its reference value.
So far, only safety indicators that aim at human health have been
considered in actual
studies in Germany. Other protection goals like protection of
non-human biota or even
the inanimate environment should be taken into account. Some of the
indicators con-
sidered in ISIBEL are of a more general character and could be
adequate for such a
concept.
Since the number of possible protection goals is nearly unlimited,
a classification of
such goals with a hierarchical structure could be a sensible task.
It should be tried to
find a limited number of protection goals that cover large ranges
of others, ideally the
total field of ‘safety’. This could help defining a limited but
comprehensive set of safety
indicators.
Safety indicators of the risk type have not been considered so far
in German studies.
The reason might be that scenario probabilities are hard to
determine. This kind of indi-
cators can, however, be very illustrative and helpful in
communicating with the public
and should therefore be tested.
Performance indicators are always helpful to better understand the
functioning of the
system. They should be defined specifically for each study. As
already seen in SPIN, it
18
is hard to give a general recommendation for the use of performance
indicators. It can,
however, be said that integrated fluxes from different
compartments, if interpreted cor-
rectly, in many cases provide very illustrative and useful
information.
2.3 Uncertainty management and uncertainty analysis
2.3.1 Background/Introduction
There are two basically different ways to handle uncertainties. One
is using conserva-
tive models and parameter values instead of realistic ones, making
sure that the reality
cannot be worse than the calculated results. The other possibility
is to establish proba-
bility distributions for all uncertain parameters and to perform a
probabilistic analysis
with a big number of separate runs. The first approach can cause
some problems as
conservativity is sometimes hard to prove. Moreover, too much
conservativity can re-
sult in a failure of the proof of safety. Probabilistic analysis is
always to be preferred, as
it allows for an assessment of the probability of a failure, as
long as the uncertainties of
models and input parameter can be properly quantified. This,
however, is not always
possible. Therefore, normally, both approaches are combined by
using conservative
models and parameters only where the uncertainty is hard to
quantify, and then per-
forming a probabilistic analysis.
Probabilistic uncertainty analysis, though not required by valid
regulations, is a com-
mon means for assessing the outcome of a repository model and has
been in use in
Germany for more than twenty years. The procedure was performed
already in 1988 in
the PAGIS study for a HLW/SF repository in rock salt and is
described in the report
/STO 88/. In later studies it was applied in the same form and
using the same tools up
to now, recently for the LAW repository near Morsleben (ERAM) and
the experimental
LAW/MAW repository in the salt mine Asse near Wolfenbüttel. The
methodology for
uncertainty assessment is approved. The main problems lie in
identifying the essential
uncertainties, finding the adequate probability distribution
functions and correct inter-
pretation of the results.
2.3.2 Key terms and concepts
In the following, the general problem of uncertainties in long-term
safety assessments
is described as it is seen by GRS (Braunschweig).
Aleatory and epistemic uncertainties
Basically, it can be distinguished between two different kinds of
uncertainties which re-
quire their specific handling: Uncertainties that are due to
physical imponderabilities or
principally unforeseeable processes are called aleatory;
uncertainties, however, that
originate from our lack of knowledge about the nature are called
epistemic. Epistemic
uncertainties are those of physical parameters that are only
insufficiently known. Such
uncertainties can be principally reduced by additional
measurements, improvement of
measurement techniques or other investigations. Aleatory
uncertainties, however, can
neither be avoided nor reduced and have simply to be accepted as
they are. An exam-
ple for an aleatory uncertainty is the time of failure of a single
canister. This depends
on things like pitting corrosion due to the existence of
microscopic fissures in the con-
tainer material from the fabrication process or from mechanical
impacts during the em-
placement. Of course, one can argue that it is possible to reduce
this uncertainty by op-
timising the canister fabrication and handling processes, but such
measures would
change the system itself and not simply the knowledge about
it.
The adequate handling of uncertainties depends on their type.
Aleatory uncertainties
should be quantified as exactly as possible and their influence on
the uncertainty of the
results should be analysed. This uncertainty has to be accepted and
taken into account
in the safety case. A sensitivity analysis normally makes little
sense for parameters that
are subject to aleatory uncertainties. In contrast to this, if
applied to epistemically un-
certain parameters, sensitivity analysis can identify those
parameters that should be
analysed or measured more thoroughly in order to reduce their
uncertainty.
In the practice of long-term safety assessments for final
repositories, there are very
few, if any at all, purely aleatory uncertainties. Most
uncertainties are a mixture of both
types, since there are random influences as well as lack of
knowledge. The epistemic
character, however, is dominant in most cases, and if it is not,
like in the mentioned ex-
ample of the canister failure time, it can nevertheless make sense
to treat the uncer-
tainty as if it were epistemic. The reason has been indicated
above: Normally, there are
possibilities to reduce even aleatory uncertainties by technical or
constructional
20
measures, and it might be helpful to identify influential
parameters by sensitivity analy-
sis. Therefore, GRS decided not to distinguish between aleatory and
epistemic uncer-
tainties and to treat all uncertainties as epistemic ones.
Kinds of uncertainties
The most important uncertainties in long-term safety assessment are
parameter uncer-
tainties. As explained above, it is always assumed that these
uncertainties are epistem-
ic, i. e. due to insufficient knowledge about the actual natural
conditions. Parameter un-
certainties can origin from poorly known properties of the host
rock, unclear flow
conditions inside the mine, lack of knowledge about chemical
conditions, etc. Parame-
ter uncertainties are relatively easy to handle because they
correspond directly with
quantifiable numerical uncertainties. In many cases, a conservative
value can be given,
but this is only possible if the influence of the parameter to the
result is monotonic.
Another kind of uncertainties is model uncertainties. In some
cases, it is not clear which
model has to be applied to describe a specific effect. Such
uncertainties can be due to
improper physical knowledge of the process, insufficient accuracy
of the available
models, or the inability to predict the correct physical situation.
Model uncertainties are
also always assumed to be epistemic. They are more difficult to
handle than parameter
uncertainties as they are hard to quantify. Where it is possible to
specify a conservative
model, this is the most convenient approach. If, however, there is
no model that can be
proved to be conservative, the model uncertainty can be mapped to
an artificial param-
eter with discrete values, each representing one of the possible
models. This parame-
ter can be treated like a normal uncertain parameter in a
probabilistic analysis.
Scenario uncertainties are the third kind of important
uncertainties in long-term safety
assessments. Normally, a number of different scenarios are
developed which are con-
sidered more or less probable. Scenarios are derived from a FEP
(features, events,
processes) analysis and comprise things like the temporal evolution
of the near field,
transport through the far field and exposition paths in the
biosphere. Since the probabil-
ities of many FEPs can only roughly be estimated, scenario
probabilities are very un-
certain. The usual method to handle these uncertainties is
investigating several scenar-
ios independently, including a worst-case scenario and a scenario
that is assumed to
represent the intended evolution. Another possibility is to
calculate risks which include
contributions from all scenarios, but this requires a proper
knowledge of the scenario
probabilities.
21
2.3.3.1 Methodology
This section describes how uncertainties have been handled within
the long-term safe-
ty assessment studies of GRS (Braunschweig). The general procedure
has been basi-
cally the same for more than 20 years. The examples in the
following are taken from
the ERAM study for the LAW repository in an abandoned salt
production mine near
Morsleben. This is one of the most recent and most detailed studies
by GRS.
Scenario uncertainties have been treated, as mentioned above, by
investigating a nor-
mal evolution scenario, a worst-case scenario, and a limited number
of additional sce-
narios that appear interesting by some reason. A quantification of
scenario probabilities
and calculation of risks has never been performed so far. Model
uncertainties have
mainly been handled by using conservative models. In some cases,
however, model al-
ternatives have been switched by use of artificial parameters as
described above. In
such cases, the model uncertainty is mapped to a parameter
uncertainty and can be
treated in the same way. Therefore, in the following only parameter
uncertainties are
considered.
Identification of uncertain parameters
Not all parameters in a safety assessment are uncertain.
Geometrical dimensions of
containers, distances in the mine building or well-known material
constants like the
mass density belong to the parameters that are more or less exactly
known. Others
may be less well-known, but are likely to have little influence on
the results and can al-
so be considered certain. In cases of doubt the value is chosen
conservatively. In the
ERAM study examples of such parameters are the void volumes in the
different levels
of the mine, or the radionuclide inventories, which have been
collected over decades
and can in some cases only be estimated.
The number of parameters that are really treated as uncertain
should be kept limited, in
order to allow a manageable uncertainty analysis. If for parameter
a clearly conserva-
tive value can be given that is not too far away from the most
probable value, are pref-
erably simply assumed to be certain. Particularly those parameters
that are suspected
to have a nonlinear or unclear influence on the calculation results
are selected for an
22
uncertainty analysis. In the ERAM study, these are 43 parameters,
comprising things
like global and local convergence rates, reference porosity,
corrosion rates, gas entry
pressure, initial permeabilities of seals, distribution
coefficients and diffusion constants.
Bandwidths and probability distribution functions
Each uncertain parameter has to be assigned a bandwidth interval.
This can be a diffi-
cult task, as, if chosen too small, the bandwidth does not come up
to the real uncertain-
ty and, if chosen too big, it could jeopardise the proof of safety.
Therefore, the interval
boundaries have to be fixed carefully and with as much expertise as
possible.
The next step is defining a probability distribution function (pdf)
for each uncertain pa-
rameter. There is no unique procedure for this task. So far, mainly
three types of distri-
butions have been used:
− Uniform distribution: If a parameter is known (or suspected) to
lie anywhere be-
tween the boundaries with no preferred value, a uniform
distribution is applied. In
some cases the interval is divided into sub-intervals with
different but constant
probabilities. This is sometimes called a histogram
distribution.
− Triangular distribution: If the parameter has a clearly preferred
value within its in-
terval but no other information is available, a triangular
distribution should be cho-
sen. It can be symmetric or asymmetric.
− Normal distribution: If a preferred value and a typical deviation
is known, a normal
distribution should be chosen. From a mathematical point of view, a
normal distri-
bution extends to infinity on both sides, which is physically
doubtful and numerically
problematic. Therefore, an interval is defined also for these
parameters and the
distribution must be cut at the boundaries. Sometimes, it seems
plausible to
choose a normal distribution within a given interval around some
mean value but
the standard deviation is unknown. In this case, the standard
deviation has to be
calculated from the interval boundaries. It is common practice to
take the bounda-
ries as the 0.001- and 0.999- quantiles of the distribution, which
corresponds to a
bandwidth of 3.09 times the standard deviation to both sides of the
mean. This is
unchangeably fixed in the EMOS code package, which has been used
for all GRS
studies. Therefore, it is neither possible to choose an asymmetric
normal distribu-
tion nor to define the interval boundaries and the standard
deviation independently.
23
All distribution types, except the triangular distribution, can be
applied either on a linear
or on a logarithmic scale. If the interval spans more than one
order of magnitude, a
logarithmical distribution is preferred. This pertains to
parameters like diffusion con-
stants, distribution coefficients or permeabilities. If the
interval is smaller than one order
of magnitude, normally a linear distribution is adequate.
Deterministic parameter variations
In the normal procedure of a safety assessment study a reference
case is defined for
each scenario under consideration. Every parameter is assigned a
reference value
within its bandwidth interval, which is either considered the most
probable value or a
slightly conservative one. The first exercise to investigate the
influence of the uncer-
tainty of a parameter is a deterministic parameter variation. The
parameter is varied be-
tween several discrete values within its bandwidth interval,
normally the boundaries
possibly a few additional values, while all other parameters are
kept on their reference
value. Comparing the results with those of the reference case and
interpreting the dif-
ferences in detail often yields valuable information about the
influence of the parame-
ter. This information, however, has a qualitative character und
must not be misinter-
preted. If the results hardly change under variation of a specific
parameter, this does
not necessarily mean that the parameter generally has little
influence. The observed
behaviour can be due to the specific situation that results from
the reference values of
the other parameters and can be totally different for another
combination of values.
The variation of a single parameter, keeping all others constant,
is called a local pa-
rameter variation. The word ‘local’ does not mean that the
variation is very small but re-
fers to the fact that only one of the parameters is
considered.
Probabilistic uncertainty analysis
For a quantitative determination of the uncertainty of the result
of a model calculation, a
probabilistic uncertainty analysis must be performed, varying all
parameters within their
bandwidths and regarding their pdfs at the same time. The model is
run for a number of
times, each with a new set of parameter values. A complete set of n
parameter value
sets is called a sample of size n.
The necessary sample size can be derived from accuracy
requirements. In Germany,
there is no official regulation so far, but criteria of 90/90,
95/95 or 99/90 are discussed.
24
The first of these numbers specifies the minimum percentage of
adherence to some
safety criterion normally given in form of a limit; the second
number is the statistical re-
liability of this statement in percent. A criterion of this type
specifies the admissible
number of limit exceedings, but does not say anything about the
acceptable amount by
which the limit is exceeded. It can be shown that, if the sample is
randomly chosen and
all calculated results remain below the limit, a sample size of 22,
59, or 230 is sufficient
to prove the 90/90, 95/95 or 99/90 criterion, respectively. This
does not depend on the
number of parameters. The actual number of runs, however, has been
essentially
higher in most studies.
There are different sampling strategies. GRS has most often used a
random sampling
strategy because it guarantees a statistical independence of the
parameter values,
which is often required by the mathematics. Intended parameter
correlations can be
taken into account as well in the sampling as in the evaluation. In
some older studies
Latin Hypercube Sampling (LHS) was applied, which allows a better
covering of the to-
tal bandwidth of each parameter.
For evaluating the results and assessing the uncertainty of a model
calculation, several
statistical measures like mean, median or maximum are calculated.
This can either be
done for the absolute maxima of all runs or a specifically
interesting point in time. If cal-
culated in small steps for the total model time, the statistical
values can be plotted as
time curves. Another curve that is valuable for the uncertainty
analysis and has always
been plotted in GRS studies is the Complementary Cumulative
Distribution Function
(CCDF). It represents the relative frequency of runs with absolute
maxima above some
value versus this value. Typically, this curve has an s-shape,
starting at 1 with a rela-
tively steep decrease in the middle region and a flat tail at the
end, finally reaching 0. It
allows a much better assessment of the adherence to some limit than
a simple statisti-
cal criterion like those mentioned above. Very useful information
can also be gained
from scatterplots with one dot for every run, each showing the
maximum value and the
time of its occurrence. These plots show, on the first sight, the
highest maxima as well
as the most critical time intervals. Additional interesting
information can be extracted if
the dots are coloured according to some properties of interest. In
the ERAM study, the
dots have been coloured after the radionuclide responsible for the
absolute maximum.
The plots show very clearly which radionuclides are responsible for
the earliest, the lat-
est, the highest, and the most maxima for each scenario.
25
Sensitivity analysis is an own topic, but since probabilistic
sensitivity analysis is closely
related with uncertainty analysis it is briefly addressed
here.
On the basis of a probabilistic set of calculations a global
sensitivity analysis can be
performed, meaning that the sensitivity of the calculation result
to individual parameters
under consideration of the influences of all others is
investigated. A sensitivity analysis
requires a much higher sample size than an uncertainty analysis. On
the other hand,
the sample size is limited by the computing time. By this reason,
in older studies the
sample size was typically a few hundred, while in the ERAM study it
was chosen to be
2 000. Generally spoken, the sensitivity analysis is the more
accurate, the bigger the
sample is.
There are a number of different methods for probabilistic
sensitivity analysis. One sim-
ple approach, named after Pearson, is to calculate the correlation
coefficients between
the output of the model and each individual input parameter. The
higher the absolute
value of the correlation coefficient is, the higher is the
sensitivity to the respective pa-
rameter. A positive coefficient means that the result increases if
the parameter does so,
a negative value indicates an inverse correlation. Another
technique is performing a
linear regression and determining regression coefficients for each
parameter. A high
regression coefficient means a high influence of the parameter to
the result. There are
some more similar, but more sophisticated, methods. All these
methods are linear,
which means that they work best for linear systems. Since, however,
the models for fi-
nal repositories are typically very complex and non-linear, the use
of these methods is
limited. A possibility of improving their significance is to
perform a rank transformation.
This means that each parameter value as well as the output value is
replaced by its
rank in the ordered list of all values in the sample. The rank
transformation makes
many models, at least monotonic ones, closer to linear, but at the
cost of losing the
quantitative relevance of the results. So far, GRS (Braunschweig)
has always per-
formed a rank transformation in sensitivity analysis studies.
A somewhat different approach to sensitivity analysis is two-sample
tests like the
Smirnov test. The sample values of the parameter under
consideration are divided in
two groups, one containing the upper 10 %, the other the rest. If
there is a significant
difference between the results obtained with the two groups, the
parameter is consid-
ered important.
tracted attention. Such methods use the statistical variance for
calculating sensitivity
measures that do not require linearity or monotonicity of the model
and can be quanti-
tatively interpreted, but need high sample sizes. The most general
theory was given by
Sobol, but the technique proposed by him is complicated and
computational expensive.
A more practical approach is the Fourier Amplitude Sensitivity Test
(FAST), which is
based on the idea to scan the parameter space periodically with
individual frequencies
for each parameter, and to recover the frequencies in the model
output value by means
of a Fourier analysis. It can be shown that the sensitivity
measures calculated with
FAST are the same as those proposed by Smirnov. The FAST method has
not yet
been applied by GRS in practical studies, but it has been tested
for demonstration pur-
poses. It could be shown that the FAST technique works and can
yield valuable addi-
tional information, compared with a linear sensitivity
analysis.
Linear as well as variance-based sensitivity analysis can be
performed with the soft-
ware tool SIMLAB which is planned to replace the statistical
components of the EMOS
package in future.
2.3.3.2 Related topics
The issue of uncertainty management is related to a number of other
topics:
− the assessment strategy,
− the safety approach,
− definition and assessment of scenarios,
− safety indicators and performance/function indicators,
− sensitivity analysis,
− modelling strategy,
27
2.3.3.3 Databases and tools
The EMOS code package used for the GRS studies automatically
calculates three lin-
ear sensitivity measures on a rank basis (Spearman rank
correlation, partial rank corre-
lation, standardised rank regression), and the Smirnov test. The
methods are applied to
the maximum value as well as to a number of points in time that may
appear interest-
ing. The parameters are ranked after the calculated significance
for each method, and
then an average ranking is calculated.
Linear as well as variance-based sensitivity analysis can be
performed with the soft-
ware tool SIMLAB which is planned to replace the statistical
components of the EMOS
package in future.
2.3.3.4 Application and experience
The results of uncertainty analysis are usually presented in
different forms. In all Ger-
man studies performed in the past, the complementary cumulated
density function
(CCDF) for the maximum was plotted. That means that the maximum
output values of
all runs, regardless of their time of occurrence, are evaluated
together. The cumulated
frequency of maxima above some value is plotted against this value.
This results typi-
cally in an s-shaped curve starting at 1 for very low output values
and ending at 0 for
very high ones. Another method of presentation is a histogram plot
directly showing the
frequencies of maxima lying in specific intervals. Both diagrams
are shown together
exemplarily for the ERAM study in figure 2.3. It can be seen that
two of 2 000 runs yield
maxima slightly above the limit.
28
2
4
6
20
40
60
80
100
Limit
Fig. 2.3 Complementary cumulative density function (CCDF) and
frequency den-
sity for the ERAM study (2 000 runs)
A very illustrative way of presenting the results of a
probabilistic analysis is shown in
figure 2.4 for the ERAM study. The absolute maxima of all runs are
plotted in a scatter
diagram versus the time of their occurrence. Additional information
is provided by col-
our-coding the radionuclides that are responsible for the
respective maxima. Only five
different radionuclides appear in the diagram. The earliest maxima
occur after a few
hundred years and are caused by 90Sr or 137Cs, which are relatively
short-lived. These
maxima are due to the extremely pessimistic assumption that the
whole mine is flooded
instantaneously after repository closure. The most maxima are
caused by 126Sn and
remain well below the limit of 3·10-4 Sv/yr. At medium times there
are some maxima
caused by 14C, at very late times 226Ra as a decay product of 238U
becomes dominant.
A few maxima at medium times are caused by 226Ra from the
inventory.
29
10-8
10-7
10-6
10-5
10-4
10-3
10-2
limit
2.3.3.5 On-going work and future evolution
It is planned to create a basis for a more systematic uncertainty
management. This
comprises unique rules for establishing appropriate probability
distribution functions
according to the degree of knowledge, as well as applying
standardised criteria for
evaluation of the results.
2.3.4 Lessons learnt
Uncertainties can be managed by using conservative models or values
or by probabil-
istic methods. Both approaches should be applied as they complement
one another. A
probabilistic uncertainty analysis should always be performed since
it is the only possi-
bility to provide quantitative measures that can be checked against
formal criteria. The
sample size has to be oriented at the formal criteria to be held,
as well as the require-
ments of the methods to be applied.
A sensitivity analysis is a very useful supplement to a pure
uncertainty analysis and
should always be performed. Deterministic parameter variations help
understanding
the system behaviour and provide a qualitative local sensitivity
analysis. A global sensi-
30
tivity analysis requires probabilistic techniques and should be
performed in combination
with the uncertainty analysis.
The methods for defining bandwidths and pdfs are not very
systematic so far. Often
they are defined by a quick expert guess. This is not satisfying.
There should be a clear
and transparent procedure which leads to a unique bandwidth and pdf
under consider-
ation of all available knowledge. The development and testing of
such a procedure is a
task of the next years.
The linear methods of sensitivity analysis, which have been applied
exclusively so far,
seem to be insufficient to analyse the system behaviour correctly.
It is possible that
they yield even misleading results. Therefore, variance-based
methods should be test-
ed in detail, the more as the computational powers of modern
hardware allow increas-
ingly big sample sizes. It has been showed that such methods can
yield some added
value. There is, however, no experience so far about necessary
sample sizes or specif-
ic problems like the considering of statistical parameter
correlations.
2.4 Evolution of the repository system
2.4.1 Background
This document describes the approaches applied by GRS-B for
analysing and imple-
menting the evolution of the repository system in performance
assessment (PA) model-
ling for the disposal of high level waste (HLW) and spent fuel in
salt rock formations in
Northern Germany. This document deals neither with the evolution of
repositories for
intermediate level waste (ILW) nor repositories for other host rock
types such as clay or
granite.
In Germany salt domes are one of the favoured options for the
disposal of waste in
deep geological formations. The large number of more than 200
existing salt domes in
Northern Germany shows, that salt masses can be exposed to
deformation and
halokinesis without being significantly dissolved, even during
times where glacial and
interglacial periods occurred and periodically covered the area.
During some glacial pe-
riods the salt formations were covered by ice sheets of several 100
m thickness (figure
2.5) exposing the formations to high mechanical stress and causing
inflow of a high
amount of freshwater into the overburden.
31
Fig. 2.5 Location of salt domes in Germany and the extension of ice
covers dur-
ing the last glacials /NOS 08/
This high persistency to mechanical stresses and other exogenic and
endogenic geo-
logical processes in the past gives a good indication that the salt
domes in Northern
Germany can provide stable conditions for deep geological
repositories (DGR) in the
future.
2.4.2 Regulatory requirements
As said above, there are no regulatory requirements or guidelines
how to deal with the
evolution of the repository system up to now. However, PA must be
performed accord-
ing to the state of the art. For a detailed description of these
criteria see the contribu-
tion of GRS-K. The safety criteria undergo currently a thorough
revision. The new regu-
lations have not been fixed yet, but it becomes apparent, that they
will contain some
requirements and statements that have a direct impact on the
assessment of the evolu-
tion of a repository system. The following issues have to be taken
into account:
32
− the main aspect of the safety concept is the proof of the safe
enclosure of the em-
placed waste; the most important component of this concept is the
proof of the in-
tegrity of the isolating rock zone,
− the assessment period is one million years,
− possible evolutions of the repository system have to be
distinguished according to
their probability; the probability defines the way how to deal with
an evolution of the
repository system and its consequences,
− the assessment of a human intrusion to the repository can not be
carried out by de-
fining probabilities; these evolutions must be analysed in a
special set of scenarios,
− events with direct consequences on human health that outreach the
consequences
of the repository system influenced by this event are not to be
regarded (e. g. im-
pact of a large meteorite),
− the nature of the biosphere and the diet of future generations
can not be predicted
for the whole assessment period; the evolution of the biosphere has
to be present-
ed in standardised or simplified way based on today’s
conditions.
2.4.3 Key terms and concepts
2.4.3.1 Safety concept
As stated in the regulatory requirements, the safety concept is
based on the proof of
the safe enclosure of the emplaced waste and its isolation from the
environment. The
proof of safety is based on numerical analyses and a collection of
plausible arguments
that support the concept for a defined safety level. The safety
level and the required
grade of isolation have not been defined yet. The main barrier is
provided by the tight
and long-term stable rock salt formation. The safety concept is
thus focussed on the
proof of the integrity of the salt formation, which is supposed to
guarantee the isolation
of the waste.
The function of the engineered barriers is to reseal the disturbed
salt rock formation af-
ter the closure of the repository and to prevent the inflow of
significant volumes of
brines into the repository until the convergence of the rock salt
seals the man-made
voids and cavities and the safe enclosure of the waste is
ensured.
33
2.4.3.2 Repository design
Since a site has not been selected yet in Germany, all design
studies have a prelimi-
nary character. However, based on the defined safety concept the
following features
must be considered for a repository in rock salt:
− sealing of shafts and access drifts,
− backfilling of voids and cavities with crushed salt,
− minimisation of the water content within the repository (e. g.
backfilling moisture) in
order to minimise container corrosion and gas production,
− limiting the maximum temperature in the rock salt formation to
200 °C in order to
avoid mineralogical or crystallographical changes of the rock
salt,
− thorough exploration of the salt formation in order to minimise
the possibility of the
occurrence of undetected brine inclusions,
− sufficient distance to brittle (e. g. anhydrite) and thermally
unstable (for tempera-
tures < 200 °C, e. g. carnallitite) salt layers as well as
adjacent rock formations,
− sufficient thickness of the salt formation above the emplaced
waste in order to min-
imise the effect of subrosion on the integrity of the geological
barrier.
2.4.4 Treatment in PA modelling
There is a high uncertainty in predicting the future development of
a repository system
over long time periods. One method to deal with this inevitable
uncertainty is the selec-
tion of a set of scenarios, which describes several possible
evolutions of the repository
system. In this method, called scenario development, a single
scenario specifies one
possible set of features, events and processes (FEP) and provides a
description of
their characteristics and sequencing /NEA 01/. In a scenario
development a set of such
scenarios must be defined and discussed that contains a complete
coverage of all rel-
evant possible future evolutions of the repository system.
The main objective of the scenario development is the
identification of relevant FEPs
that affect the future behaviour of the repository system and the
synthesis of these
FEPs to an appropriate set of scenarios (i. e. calculation cases
for PA models). Beside
its importance for the scope of the PA modelling procedure scenario
development is
34
essential for the communication of the modelling results and its
underlying assumptions
to the public. For this reason the scenario development has to be
as systematically and
transparently as possible.
In the past two basic approaches have been applied in
Germany:
− the identification of all FEPs that can have an influence on the
repository system
and the emplaced waste and development of scenarios by combining
these FEPs
to plausible scenarios (bottom-up approach).
− the determination of initiating FEPs for scenarios, in which
barrier functions in the
repository system are affected in such a way that a contact between
brine and
waste is possible, and identification of FEPs that are relevant for
these scenarios
(top-down approach).
The first approach has the advantage to be more objective and
traceable, but the step
from a complete FEP-list to a set of scenarios has not been
accomplished yet without
using elements of the top-down approach. For the salt domes in
Northern Germany a
FEP-list for spent fuel and HLW was generated exemplarily for a
reference site taking
into account both approaches /DBE 08/. The FEP-database developed
by OECD/NEA
/NEA 00/ provided the starting point for this FEP-list.
Currently this list is used for the definition of a complete set of
scenarios. This work has
not been completed yet, but it is commonly accepted (see WP1.1
‘scenario develop-
ment’), that such a definition of scenarios must contain a
definition of the expected de-
velopment of the repository system, the normal evolution scenario.
All other probable
and less probable developments must be defined in altered evolution
scenarios. The
distinction between probable and less probable evolutions must be
carried out accord-
ing to the regulatory requirements.
2.4.4.1 Normal evolution scenario
The safety concept discussed in chapter 3.1 requires a new
definition of the normal
evolution scenario for a salt dome in Northern Germany. This
definition has not been
carried out yet, but it should be based on the following
assumptions:
− there are no transport pathways in the host rock (the integrity
of the host rock has
been proven),
− all geotechnical barriers fulfil their functions during their
designed lifetime,
− the material for the backfill and the seal can be compacted in a
way, that its re-
maining permeability is low enough to ensure the isolation of the
waste from the
groundwater,
− accumulated gas can penetrate the host rock without impairing its
integrity,
− the maximum rock temperature will not exceed 200 °C.
In order to make the scenario approach more structured the evidence
period of the
normal evolution scenario is generally divided in several
sub-periods. Normally two
main phases are distinguished.
Thermal phase
The thermal phase is defined as the time period where the heat
generated by the em-
placed waste has a relevant impact on the temperature in the salt
formation. Depend-
ent on the definition of a relevant thermal impact this period ends
between 103 and 104
years after the closure of the repository. According to the
assumptions given above the
convergence of the salt will produce a complete consolidation of
the backfill material
within several centuries in the normal evolution scenario. As a
consequence the em-
placed waste will be isolated within the rock salt formation and no
radionuclide release
from the repository will occur. An excerpt of FEPs from the exi